Application of machine learning and laser optical-acoustic spectroscopy to study the profile of exhaled air volatile markers of acute myocardial infarction

Conventional acute myocardial infarction (AMI) diagnosis is quite accurate and has proved its effectiveness. However, despite this, discovering more operative methods of this disease detection is underway. From this point of view, the application of exhaled air analysis for a similar diagnosis is valuable. The aim of the paper is to research effective machine learning algorithms for the predictive model for AMI diagnosis constructing, using exhaled air spectral data. The target group included 30 patients with primary myocardial infarction. The control group included 42 healthy volunteers. The 'LaserBreeze' laser gas analyzer (Special Technologies Ltd, Russia), based on the dual-channel resonant photoacoustic detector cell and optical parametric oscillator as the laser source, had been used. The pattern recognition approach was applied in the same manner for the set of extracted concentrations of AMI volatile markers and the set of absorption coefficients in a most informative spectral range 2.900 +/- 0.125 mu m. The created predictive model based on the set of absorption coefficients provided 0.86 of the mean values of both the sensitivity and specificity when linear support vector machine (SVM) combined with principal component analysis was used. The created predictive model based on using six volatile AMI markers (C5H12, N2O, NO2, C2H4, CO, CO2) provided 0.82 and 0.93 of the mean values of the sensitivity and specificity, respectively, when linear SVM was used.

Automated summary

This study utilized exhaled air analysis and machine learning algorithms to construct a predictive model for AMI diagnosis, showing that under certain conditions, high sensitivity and specificity can be achieved.